|Solid State Talk all about solid state amplification.|
Please consider donating to help us continue to serve you.
Ads on/off / Custom Title / More PMs / More album space / Advanced printing & mass image saving
||Thread Tools||Search this Thread|
|17th April 2019, 08:31 PM||#1|
Join Date: Sep 2006
Easy-MOS is a simplistic, efficient and evolutive all-MOS amplifier for beginners
This amplifier is terminally simple, cheap and forgiving, but it remains capable of amplifying in the seventies style: it would have been capable of meeting some HiFi standards of the time.
Of course, nowadays a THD of 0.5% and a full-power bandwidth of 50kHz is only tolerable in audiophile or tube amps, but even compared to a chipamp, it remains easy and accessible.
The design came to birth as a by-product (understand reject) of a request from Daniel: he asked me to design a discrete amplifier that could match a chipamp in performance, simplicity and ease of construction.
I examined a number of options, and I rejected this one because of its inverting design and its (relatively) low performances: in its best implementation, it manages 0.1% THD and 100kHz power bandwidth.
It is nevertheless a neat and effective amplifier, and importantly, it is highly evolutive: this is the plain vanilla version, but serious improvements are possible without too many complications; more on that later.
The main drawback of ultra-simple MOS-followers based amps is their output voltage swing heavily curtailed by the MOS threshold voltage, typically several volts under load.
Add the inevitable losses from the source resistors and the drivers and you are left with an output of Vrail minus 5 to 10V for each rail.
For high-power amps, the loss is relatively benign, but with 10 or 15V of total supply voltage, practically nothing is left to drive the load.
EZ-MOS solves the problem by bootstrapping for the positive rail and a "bias memory cap" for the negative one (or GND for a simple amplifier like this one).
The result is a 3-transistor + bias topology capable of swinging 28.5Vpp with a 31V supply: not bad for a simple follower-based amp.
One of the main drawback is the inverting configuration, but even renowned designers like Renardson use it.
Here are pics of my prototype:
The low-power MOSfets can be almost any low-power type like 2N7000, and the OP can be any approximate complementary vertical pair.
My prototype is wired for a ~30V supply and is capable of putting out ~12W @ 8Ω, but other voltages and load impedances are possible just by scaling R13.
You can probably notice that the board is heavily populated compared to the schematic: that's because of the evolutive aspect.
The prototype includes all the options, but here the jumpers are in the "plain vanilla" position.
More will follow
Here is the 10kHz, large signal squarewave response:
Not blindingly fast for this version obviously, but clean and sufficient for audio purposes, at least in principle.
The simulated THD is a shade over 0.5%; in the same conditions, the measured THD is a shade under 0.5%, thus very close, which is normal for "high" THD levels.
|18th April 2019, 12:30 AM||#3|
Join Date: Dec 2014
Elvee, that looks very simple and interesting!
Any plans for symmetrical +/- rails, to get rid of the output cap?
Also, what other output devices can be used?
|18th April 2019, 01:38 AM||#4|
Join Date: Apr 2010
Location: Coffs Harbour, NSW
Ahhh...my kind of experimental DIY I like it, whether it is 0.01, 0.1 or even 1% THD but the question is, what would be the approximate or typical ratio of harmonics in the distortion in this design and as you follow through to other implementations? This feature is really significant among popular old designs and also to what we do now, because it determines whether we can live with small amounts of distortion or not. That makes it a good predictor of popularity for the longer term.
If only I had listened....
|18th April 2019, 12:28 PM||#6|
Join Date: Mar 2011
Location: france (bretagne)
i like that i see Elvee .
which is the most important ?, the irf530 or the rfp15p05?
it is a mandatory choice or there is room for maneuver in the choice of output transistors?
|18th April 2019, 04:35 PM||#7|
I suspect D1 tries to prevent C3 from pushing current into the +30V top supply rail on a high amplitude peak. With D1 in place the charge on the "top plate" of C3 can be leaked away through a relatively high resistance (>30K) path that ends in R11. Without D1 it can be leaked away through only 470 ohms (R1).
|18th April 2019, 05:30 PM||#8|
Join Date: Sep 2006
That said, adding a transistor in the seventies style would be perfectly possible, but this would become another amplifier.
I may discuss it later
If the bias adjustment range becomes insufficient, you can swap the 100K trimpot for a 220K one.
About the bias, 200~250mA is a good trade-of between heat and low-level linearity, but you are free to chose different values
Because of the non-symetrical topology and the low feedback, low-order, even harmonics tend to be dominant.
Here is the THD residue (0.15%) together with the output signal:
When the potential on the positive armature of C3 exceeds the +rail, it prevents its discharge through R1, thus reducing the rms current through this resistor and the subsequent heating, and it also allows C3 to charge at a higher voltage than the normal Vs/4 value, allowing a larger bias current through the VAS and the bias spreader.
Of course, MOSfets don't require a base current, but they need some capacitive current under high slewrate conditions, and this little trick helps a little.
This "dynamic bias" feature can be tweaked by changing the ratio of R1 to R2 (or it can be eliminated if you don't like it)
The first and most obvious upgrade to the base circuit is to add a cascode stage: because of the inverting topology, R12 and its dependent resistors R10 and R13 need to be reasonably large; combined with the largish MOS capacitances, this results in a strong Miller effect.
This effect self-compensates the amplifier, but it is much larger than necessary, resulting in poor dynamic performances.
The solution is extremely simple: add a properly biased device in M3's drain.
The device can be a MOS, BJT or jFET.
A MOS works, but due to its finite transconductance and parasitic capacitances, it adds unwanted poles to the loop, resulting in instability.
The cure is easy: to add a capacitor between its drain and the input, but it kind of negates the usefulness of the cascode.
BJT's or jFETs do not require anything to work properly, but the Idss/Vto of jFETs varies widely, even for a given type, and this could require bias adjustments.
For these reasons, I recommend using a simple BJT: it is completely predictable and the bias source is an ordinary red LED.
With the same LED, a FET like the 2SK125 also works. Some high Idss types could work with no bias at all.
A good point worth noting: the addition of the cascode does not reduce the output swing.
Here is the circuit configured for a BJT cascode:
And here is the same 10kHz squarewave (note the improvement):
Here is the clipping behavior:
Another addition to the base circuit is the RC network C8/R14: it probably looks intriguing.
Its effect is to reduce the THD, from 0.4% to less than 0.15% (in the real circuit), about 10dB.
How does work?
Well, that is not entirely clear at the moment, even for me, and I still need to think about it.
I think it operates through two mechanisms: it somewhat compensates for the distortion the bootstrap reinjects at the input of the MOS (the bootstrap cap is tied to the source of the NMOS, and that's where it should be) and it forces a local degeneration of the OP's by changing the pure input current drive to a ~voltage drive, making the degeneration by the source resistors possible.
It is unexpected though that a local degeneration (reducing the overall loop gain) manages to reduce the global non-linearities: the general rule is exactly the opposite (which does not mean that local degeneration is useless, I hurry to say).
The effect is not an artifact, it has been confirmed in other simulations (this one does not show it, because I used models "similar" to the real circuit, but I include another one, less faithful but in fact more accurate), and in the real world.
If someone has an illumination, please share it....
Now, prepare yourself and fasten your seat-belts, because the next installment will take you to the other side of the quiescent current auto-bias mirror, and its simplicity will match that of the amplifier (1 transistor)!
|19th April 2019, 03:39 PM||#9|
Join Date: Nov 2010
Hi Elvee, I'd like to share a bit of experience. I built a mini F5 clone with IRF 530/9630 in 2014 with bipolar 12V supply without any source resistor on the mosfets. The circuit was thermally stable biased at ~ 450mA. I think you can also reduce the source resistor to < 0.11 ohm or even omit them altogether to gain the advantage of lower H3 and H5.
|21st April 2019, 06:41 PM||#10|
Join Date: Sep 2006
Thanks for your constructive remarks.
We have to remember that the F5 and EZmos are two very different amplifiers:
The F5 has a common-source output topology, and operates in class A, if I am not mistaken.
EZmos has a more traditional follower OP, and is primarily intended for class AB work (nothing prevents you from increasing the bias to pure class A, but that was not my initial goal).
In principle, one should compare apples to apples, or in this case prevent ourselves from applying apple recipes to oranges.
Despite these caveats, your remarks remain broadly applicable to this amplifier: it would be possible to operate with lower, or even no source resistors at all.
MOSfets, even vertical ones have some in-built compliance, unlike BJT's (under some peculiar circumstances, even BJT's can work without resistors, but that's not something I recommend to beginners or the faint of heart).
Increasing the quiescent current to 450mA would be beneficial: I have observed that the linearity improves with increasing current (in sims, there is a sweet spot at an intermediate value, but not in reality).
A reduction of the source resistors should bring apparent advantages: reduced output impedance, increased damping factor, larger output under load (here, the difference between the 28.5Vpp output and the 31V supply is mostly wasted in the resistors).
Another effect is that it unmasks the true nature of the OP devices: N- and P-MOS are fundamentally different, even when you try to match them better by using a IRF530 and a IRF9630 instead of a 9530.
Asymmetries result in the dominance of even harmonics over odd ones, and in a low-feedback amplifier such as this one, the effect remains present in the output signal.
With source resistors, the characteristics of the two initially dissimilar MOS is evened, allowing non-linearity compensation.
This compensation allows a reduction in the overall THD figure, but it will do so by eliminating the even harmonics, so cherished by some (don't count me in).
Eliminating the source resistors and increasing the bias current will improve the global linearity and favor H2 H4 against H3 H5, but I strongly advise against it:
This amplifier is a beginner's project, and to apply these measures without too many risks requires minimal competences to achieve a good thermal contact between the transistors and the heatsink.
An amateurish attachment might be enough to sustain the permanent 5 ~ 7W caused by the quiescent current, but as soon as a "thermal event" caused by the program contents and the load occurs, it will drive the amp into a fast and irrecoverable thermal runaway.
With reasonably sized resistors, this risk is mitigated.
There is a safer solution for those wanting to alter the THD profile: simply use dissimilar source resistor, for example 0 for one or 0.5 for the other.
It will have a similar effect, but without the risks associated with no resistor at all.
You could use the 0.5 one for the P side, to reinforce its "weakness", or for the N side, to get more subtle effects.
Now, let us examine the quiescent current servo:
It is built around Q2 and D4: D4 and the B-E junction of Q2 form a kind of diode gate, but with simplified idealized diodes, no current could ever flow into Q2's base: D4's anode is always at a more positive potential than Q2's base.
Actual semiconductor junctions have an exponential behavior though, and R17 in series with D4 also helps shifting the level.
The result is a "sampling gate" that conducts for each minimum of the total voltage across R3 and R4, ie. the zero-crossing of the output current waveform.
This means that the conduction of Q2 will be modulated by how low the cross-conduction (or quiescent) current is.
The collector current is integrated by C9 and alters the conduction of the bias spreader M4 via a resistive divider.
In short, the system imposes a minimal current through the output stage.
Because the amplifier is AC-coupled, the quiescent current is updated at least a few tens of time per second (or continuously under no-signal conditions).
The bias voltage is integrated and held with a time constant of several seconds by C5.
Note that since the circuit does not operate in real-time, it does not allow non-switching operation, unlike the Circlophone for example.
To make the auto-bias active, the bias trimpot is set to zero, and X2 is inserted, allowing the AB-circuit to take control.
The circuit has a very low loop gain, meaning the target current value can still be adjusted with the trimmer.
Alternatively, R9 could be made fixed and larger, and the current could be set by varying R17.
I said earlier that the circuit was not perfect: with this arrangement, the dynamic quiescent current varies between quiescent and large signal conditions.
The peak base current in Q2 increases under large signal, but not enough to compensate the reduced duty cycle, resulting in a lower average current.
That is caused by the inverse exponential behavior of the diode gate.
For example, if the static current is set to 250mA, the dynamic large signal current is reduced to ~175mA.
I prefer to see that as a "feature", not a negative: it increases the first-watt, pure class A operation range and equalizes the heat output for different conditions.
In fact, my first implementation of the circuit largely compensated the variation, but in the end I opted for this one.
The first version had the diode in the base instead of the emitter, and the bias resistor R16 was tied to the positive armature of C3; this made the average bias current dependent on the signal level, and could be made to compensate the reduction caused by the dynamic operation.
In the end, I opted for this version, also because it gives Q2 one more Vbe of working room, increasing the range of correction possible.
Here is a pic of the circuit with the AB active (not much has changed):
|Thread Tools||Search this Thread|
|Thread||Thread Starter||Forum||Replies||Last Post|
|QUASI Amplifier for Beginners||apexaudio||Solid State||891||19th November 2018 08:08 AM|
|Efficient Subwoofer Amplifier||moschfet||Solid State||21||25th November 2012 04:13 PM|
|OLD THREAD Simplistic Mosfet HV Shunt (The Simpler Simplistic Design by Salas)||quanghao||Quanghao Audio Design||460||6th May 2012 11:13 PM|
|PCB for QUASI Amplifier for Beginners with changes!?||radio||Solid State||4||18th August 2011 08:59 PM|
|New To Site?||Need Help?|